Let's say you have 10 balls of three different colours in a bag: 2 red, 3 blue, 5 green. The probabilities of randomly selecting a red, blue or green ball are 0.2, 0.3 and 0.5 respectively. Simple mathematics enables us to work out the probabilities. The mathematical probabilities are also known as theoretical probabilities.
Now consider a number of trials, say, 100 trials. In each trial, one ball is drawn. A record is kept of the results: the number of times a red, blue or green ball is selected. The expected results (predicted mathematically or theoretically) would be 20, 30 and 50. Experimentally, the results may be a little different and that would give different values for the probabilities.
Consider 1000 trials. Which set of probabilities would you use to predict the net results? You should use the mathematically derived set because it's logically more reliable. The results should show less deviation from expectation when the theoretical probabilities are used than if the experimental ones were used.